Deep learning for very short term solar irradiation forecasting

The use of renewable energy sources (RES) has increased significantly in recent years, in particular, photovoltaic energy which is one of the RES most used for electricity production. Indeed, the world has experienced the installation of a huge number of photovoltaic systems, autonomous or connected to the electricity distribution grid. However, the improvisational nature of solar energy negatively influences the stability and reliability of the electricity grid. One of the best solutions to stabilize and secure the operation of the network is to forecast energy production and to promote the integration of photovoltaic energy on a large scale.In this context, this work aims to develop appropriate forecasting models in the forecasting of photovoltaic energy production. For this reason, we have tested machine learning and deep learning techniques to predict time series data for solar irradiation. For this, we used data from preprocessing, training and testing, three types of error metrics are used to evaluate the models. Recurrent neuron network (RNN) as a machine learning model, on the other hand LSTM and GRU as deep learning models have been discussed from the mathematical and practical simulation using the Anaconda and python environment with their math libraries.